xiaoxia-saas/tests/unit/test_render_metrics.py

"""渲染引擎灰度观测指标单元测试。"""

from __future__ import annotations

import subprocess
import time

import pytest
from prometheus_client import REGISTRY, CollectorRegistry, Counter, Gauge, Histogram


@pytest.fixture(autouse=True)
def _isolate_registry(monkeypatch):
    """每个测试使用独立的 CollectorRegistry，避免互相影响。"""
    from video_processing import render_metrics

    test_registry = CollectorRegistry()

    # 用测试 registry 重新创建所有指标
    task_total = Counter(
        "render_task_total",
        "Total number of render tasks",
        ["engine", "task_type", "status"],
        registry=test_registry,
    )
    task_duration = Histogram(
        "render_task_duration_seconds",
        "Render task duration in seconds",
        ["engine", "task_type"],
        buckets=(1, 5, 10, 30, 60, 120, 300, 600, 900, 1800, 3600, 7200),
        registry=test_registry,
    )
    error_total = Counter(
        "render_error_total",
        "Total number of render errors",
        ["engine", "task_type", "error_type"],
        registry=test_registry,
    )
    stream_copy_total = Counter(
        "render_stream_copy_total",
        "Total number of stream copy attempts and results",
        ["result", "reason"],
        registry=test_registry,
    )
    active_tasks = Gauge(
        "render_active_tasks",
        "Number of render tasks currently in progress",
        ["engine", "task_type"],
        registry=test_registry,
    )

    # 替换模块中的指标对象
    monkeypatch.setattr(render_metrics, "RENDER_TASK_TOTAL", task_total)
    monkeypatch.setattr(render_metrics, "RENDER_TASK_DURATION_SECONDS", task_duration)
    monkeypatch.setattr(render_metrics, "RENDER_ERROR_TOTAL", error_total)
    monkeypatch.setattr(render_metrics, "RENDER_STREAM_COPY_TOTAL", stream_copy_total)
    monkeypatch.setattr(render_metrics, "RENDER_ACTIVE_TASKS", active_tasks)

    yield test_registry


def _counter_value(counter: Counter, **labels) -> float:
    """获取 Counter 指定标签的当前值。"""
    return counter.labels(**labels)._value.get()


def _gauge_value(gauge: Gauge, **labels) -> float:
    """获取 Gauge 指定标签的当前值。"""
    return gauge.labels(**labels)._value.get()


def _histogram_sum(histogram: Histogram, **labels) -> float:
    """获取 Histogram 指定标签的观测总和。"""
    return histogram.labels(**labels)._sum.get()


def test_render_task_metrics_success():
    """成功任务应记录 success 状态 + 耗时 + 活跃数正确增减。"""
    from video_processing.render_metrics import (
        RENDER_ACTIVE_TASKS,
        RENDER_TASK_DURATION_SECONDS,
        RENDER_TASK_TOTAL,
        render_task_metrics,
    )

    assert _counter_value(RENDER_TASK_TOTAL, engine="unified", task_type="compose_video", status="success") == 0
    assert _gauge_value(RENDER_ACTIVE_TASKS, engine="unified", task_type="compose_video") == 0

    with render_task_metrics(engine="unified", task_type="compose_video"):
        assert _gauge_value(RENDER_ACTIVE_TASKS, engine="unified", task_type="compose_video") == 1
        time.sleep(0.01)

    assert _gauge_value(RENDER_ACTIVE_TASKS, engine="unified", task_type="compose_video") == 0
    assert _counter_value(RENDER_TASK_TOTAL, engine="unified", task_type="compose_video", status="success") == 1
    assert _counter_value(RENDER_TASK_TOTAL, engine="unified", task_type="compose_video", status="failure") == 0
    assert _histogram_sum(RENDER_TASK_DURATION_SECONDS, engine="unified", task_type="compose_video") > 0


def test_render_task_metrics_failure():
    """失败任务应记录 failure 状态 + 对应错误类型。"""
    from video_processing.render_metrics import (
        RENDER_ERROR_TOTAL,
        RENDER_TASK_TOTAL,
        render_task_metrics,
    )

    with pytest.raises(ValueError, match="test error"):
        with render_task_metrics(engine="legacy", task_type="edit_plan"):
            raise ValueError("test error")

    assert _counter_value(RENDER_TASK_TOTAL, engine="legacy", task_type="edit_plan", status="failure") == 1
    assert _counter_value(RENDER_TASK_TOTAL, engine="legacy", task_type="edit_plan", status="success") == 0
    assert _counter_value(RENDER_ERROR_TOTAL, engine="legacy", task_type="edit_plan", error_type="validation") == 1


def test_render_task_metrics_ffmpeg_error():
    """FFmpeg 执行失败应分类为 ffmpeg_error。"""
    from video_processing.render_metrics import RENDER_ERROR_TOTAL, render_task_metrics

    with pytest.raises(subprocess.CalledProcessError):
        with render_task_metrics(engine="unified", task_type="generate_video"):
            raise subprocess.CalledProcessError(returncode=1, cmd=["ffmpeg"])

    assert _counter_value(RENDER_ERROR_TOTAL, engine="unified", task_type="generate_video", error_type="ffmpeg_error") == 1


def test_render_task_metrics_timeout():
    """超时异常应分类为 timeout。"""
    from video_processing.render_metrics import RENDER_ERROR_TOTAL, render_task_metrics

    with pytest.raises(subprocess.TimeoutExpired):
        with render_task_metrics(engine="unified", task_type="compose_video"):
            raise subprocess.TimeoutExpired(cmd=["ffmpeg"], timeout=10)

    assert _counter_value(RENDER_ERROR_TOTAL, engine="unified", task_type="compose_video", error_type="timeout") == 1


def test_render_task_metrics_active_tasks_decrement_on_error():
    """任务异常退出时，活跃任务数也应正确递减。"""
    from video_processing.render_metrics import RENDER_ACTIVE_TASKS, render_task_metrics

    assert _gauge_value(RENDER_ACTIVE_TASKS, engine="legacy", task_type="compose_video") == 0

    with pytest.raises(RuntimeError):
        with render_task_metrics(engine="legacy", task_type="compose_video"):
            raise RuntimeError("boom")

    assert _gauge_value(RENDER_ACTIVE_TASKS, engine="legacy", task_type="compose_video") == 0


def test_record_stream_copy():
    """stream_copy 命中/跳过/回退应分别计数。"""
    from video_processing.render_metrics import RENDER_STREAM_COPY_TOTAL, record_stream_copy

    assert _counter_value(RENDER_STREAM_COPY_TOTAL, result="hit", reason="成功") == 0
    assert _counter_value(RENDER_STREAM_COPY_TOTAL, result="miss", reason="编码不匹配") == 0

    record_stream_copy("hit", "成功")
    record_stream_copy("hit", "成功")
    record_stream_copy("miss", "编码不匹配")
    record_stream_copy("fallback", "ffmpeg失败")

    assert _counter_value(RENDER_STREAM_COPY_TOTAL, result="hit", reason="成功") == 2
    assert _counter_value(RENDER_STREAM_COPY_TOTAL, result="miss", reason="编码不匹配") == 1
    assert _counter_value(RENDER_STREAM_COPY_TOTAL, result="fallback", reason="ffmpeg失败") == 1


def test_classify_error_unknown():
    """未知异常应分类为 unknown。"""
    from video_processing.render_metrics import _classify_error

    assert _classify_error(RuntimeError("something")) == "unknown"


def test_classify_error_io():
    """IO 异常应分类为 io_error。"""
    from video_processing.render_metrics import _classify_error

    assert _classify_error(OSError("disk full")) == "io_error"


def test_engine_and_task_type_labels():
    """不同 engine 和 task_type 应独立计数。"""
    from video_processing.render_metrics import RENDER_TASK_TOTAL, render_task_metrics

    with render_task_metrics(engine="unified", task_type="compose_video"):
        pass
    with render_task_metrics(engine="legacy", task_type="compose_video"):
        pass
    with render_task_metrics(engine="unified", task_type="edit_plan"):
        pass

    assert _counter_value(RENDER_TASK_TOTAL, engine="unified", task_type="compose_video", status="success") == 1
    assert _counter_value(RENDER_TASK_TOTAL, engine="legacy", task_type="compose_video", status="success") == 1
    assert _counter_value(RENDER_TASK_TOTAL, engine="unified", task_type="edit_plan", status="success") == 1